JP2001175809A - Character reader, reading method, and reading program recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To accurately read characters even if a relatively large stain such an unerased character left when a character written with a pencil is erased with an eraser is present. SOLUTION: A character density decision part 3113 decides a character density section on a multi-valued image obtained by inputting a character image to be read on a document. A stain decision threshold calculation part 3114 calculates a stain decision threshold for discriminating an entry character to be read on the multi-valued document and a stain such as an unerased character left when the entry character is erased with an eraser. A stain decision part 3121 decides which of a character stroke and a stain connecting block pixels on the binary image are and removes the connecting black pixels judged to be a stain as noise.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character reading device, a character reading method, and a recording medium on which a character reading program is recorded, and more particularly, to removing stains such as unerased characters generated when characters written with a pencil are erased with an eraser. The present invention relates to a character reading device having a noise removing process, a character reading method, and a recording medium on which a character reading program is recorded.

[0002]

2. Description of the Related Art In a character reading apparatus, a multivalued image of a character to be read on a form is binarized by a binarization threshold value determined by some method, and character recognition is performed on the binarized image. Generally, processing is performed. However, when there are stains and the like around the character, when binarizing the multi-valued image, they may appear as noise on the binary image together with the character. As a noise removal process for such a binary image, a process of detecting an isolated connected black pixel of a predetermined size or less and removing it as noise is general. By this processing, minute noise around the character can be removed. However, in this process,
It was not possible to remove relatively large stains, such as unerased characters, which are generated when characters written with a pencil are erased with an eraser.

[0003] Conventional techniques for removing such relatively large stains are disclosed in JP-A-6-274693 and JP-A-5-6461.

In Japanese Patent Application Laid-Open No. 6-274693, a linear region is set at the position of a character on a character image, and the connected linear pixel is used as seed information to scan the linear region. Then, an operation of extracting a black pixel connected to the seed information is performed. That is, focusing on the connectivity of black pixels constituting a character, isolated relatively large connected black pixels are determined to be noise such as dirt, and are removed.

In Japanese Patent Laid-Open Publication No. Hei 5-6461, a maximum pixel block is extracted as a character from a compressed image obtained by compressing a character image by half, and after expanding this maximum pixel block to the original image size again, By taking the AND of this and the original character image, relatively large connected black pixels are removed as noise.

[0006]

However, Japanese Patent Laid-Open No. 6-2 / 1994
In Japanese Patent Application Laid-Open No. 74693, it is determined whether a character is a character or a noise based on the assumption that the character is represented by one connected black pixel. It is difficult to apply because characters represented by separate strokes can easily occur.

In Japanese Patent Laid-Open Publication No. Hei 5-6461, characters and noise are determined inside or outside a rectangular area called a maximum pixel block. Therefore, when one character is represented by a plurality of separate strokes. There is a problem that it is difficult to correctly determine a character stroke and a dirt when the dirt exists in the vicinity of a character or a character.

[0008] Therefore, in the conventional technology as described above,
Such noises have caused misreading or unreading of characters because stains such as unerased parts due to an eraser cannot be removed properly.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned drawbacks of the prior art, and has as its object to remove noise such as unerased residue generated when a character is erased with an eraser when writing with a pencil as noise. Accordingly, it is an object of the present invention to provide a character reading device, a character reading method, and a recording medium on which a character reading program is recorded, which can accurately read characters even if there is dirt or the like.

[0010]

SUMMARY OF THE INVENTION A character reading apparatus according to the present invention is a character reading apparatus for reading handwritten characters written on a form, wherein the character is read on a multi-valued image obtained by reading an image on the form. A character density determination unit that determines a density category that is a category of the density; and a dirt determination threshold that calculates a dirt determination threshold for determining a character to be read in the multi-valued image and a dirt other than the character. It is characterized by including a threshold value calculation unit and a stain determination unit that determines whether a connected black pixel on the binary image obtained by binarizing the multi-valued image is a stain or a character stroke. Further, the stain is a stain such as an unerased residue generated when erased with an eraser, and the stain determination unit determines the connected black pixel as a stain based on a pixel value of the multi-valued image. I do.

[0011] A character reading method according to the present invention is a character reading method for reading handwritten characters written on a form,
A character density determination step of determining a density category, which is a density category, of a character on a multi-valued image obtained by reading an image on the form; and a character to be read in the multi-valued image and a character other than the character. A dirt determination threshold value calculating step of calculating a dirt determination threshold value for determining dirt; and determining whether a connected black pixel on the binary image obtained by binarizing the multi-valued image is dirt or a character stroke. And a dirt determination step. In addition, the stain is a stain such as an unerased residue generated when the stain is erased with an eraser, and in the stain determination step, the connected black pixel is determined as a stain based on a pixel value of the multi-valued image. And

[0012] A recording medium according to the present invention is a recording medium in which a computer reads a character reading program for reading handwritten characters written on a form, and the program reads an image from the form with the computer. And a character density determination step of determining a density category, which is a density category, of a character on the multi-valued image, and determining a character to be read in the multi-valued image and stains other than the character. A dirt determining threshold value calculating step of calculating a dirt determining threshold value; and a dirt determining step of determining whether a connected black pixel on the binary image obtained by binarizing the multi-valued image is a dirt or a character stroke. It is characterized by. In addition, the stain is a stain such as an unerased residue generated when the stain is erased with an eraser, and in the stain determination step, the connected black pixel is determined as a stain based on a pixel value of the multi-valued image. And

[0013]

Next, embodiments of the present invention will be described in detail with reference to the drawings. Referring to FIG.
An embodiment of the present invention includes an image input device 1 such as a scanner that takes in a character image to be read on a form as a multi-valued image, a storage device 2 that stores information, and a data processing device 3 that operates under program control. And an output device 4 such as a display device.

The storage device 2 includes an image storage unit 21 and a character recognition dictionary unit 22. The image storage unit 21 stores a multi-value image of a character captured by the image input device. The character recognition dictionary unit 22 stores a dictionary used for character recognition processing.

The data processing device 3 includes a pre-processing unit 31 for performing pre-processing for character recognition, and a feature extracting unit 32 for extracting a feature for character recognition from the binary image generated by the pre-processing unit 31. And a character recognition unit 33 that performs character recognition by comparing the features extracted by the feature extraction unit 32 with the character features stored in the character recognition dictionary unit 22.

The pre-processing unit 31 includes a binarizing unit 311 and a noise removing unit 312. Binarization unit 31
1 includes a density histogram generation unit 3111, a binarization threshold value calculation unit 3112, a character density determination unit 3113, a stain determination threshold value calculation unit 3114, and a binary image generation unit 3115.

The density histogram generation unit 3111 generates a density histogram in which the number of pixels for each pixel value of the multivalued image stored in the image storage unit 21 is counted. The binarization threshold calculator 3112 calculates a binarization threshold for binarizing the multi-valued image based on the density histogram. The character density determination unit 3113 determines a character density division on the multi-valued image based on the density histogram. The dirt discrimination threshold calculator 3114 calculates a dirt discrimination threshold for discriminating between dirt and characters based on the density histogram. The binary image generation unit 3115 generates a binary image obtained by binarizing the multi-value image using the binary threshold.

The noise removing unit 312 includes a stain determining unit 312
1 and a connected black pixel removing unit 3122. The stain determination unit 3121 determines whether the connected black pixel on the binary image is stain or a character stroke. The connected black pixel removing unit 3122 removes, from the binary image, the connected black pixels determined as dirty by the stain determining unit 3121 or the connected black pixels having a predetermined size or less.

Next, in addition to FIG. 1, the overall operation of the present embodiment will be described in detail with reference to the flowcharts of FIGS.

Referring to FIG. 2, first, a character image to be read on a form by the image input device 1 is a multi-valued image I.
The image is captured as MG and stored in the image storage unit 21 (step S1). In the multi-valued image IMG, the larger the pixel value is, the brighter the pixel is.

Next, the binarizing section 311 outputs the multi-valued image IMG
To generate a binary image IMB (step S
2). In this case, as shown in FIG. 3, first, the density histogram generation unit 3111
, A density histogram is generated by counting the number of pixels for each pixel value (step S20 in FIG. 3).

Next, the binarization threshold value calculation unit 3112
A threshold value BTH for binarizing the multi-valued image IMG is calculated based on the density histogram (step S2).
1). The binarization threshold value BTH is described in 1980, IEICE Transactions Vol. J63-D, No. 4, pp. 34
In 9-356, it is determined by the discriminant analysis method proposed by Otsu.
The method proposed by Otsu is, for example, the use of text and background.
This is a method of selecting a threshold value in an image including two categories so as to minimize the variance within each category and maximize the variance between categories.

The binarization of the multi-valued image IMG is performed as follows. If the pixel value of a certain pixel of the multi-valued image IMG is smaller than the binarization threshold value BTH, the pixel value of the pixel on the binary image IMB corresponding to the pixel is set to “1”, that is, “black”, and the multi-valued image IMG If the pixel value of a certain pixel is equal to or greater than the binarization threshold BTH, the binary image I corresponding to that pixel
The pixel value of the pixel on the MB is “0”, that is, “white”.

Next, the character density determination unit 3113 determines the density classification of the character on the multi-valued image IMG (step S2).
2). The character density categories are 'blank', 'light',
It is assumed that there are three sections of “dark”.

As shown in FIG. 4, the character density determining section 3113 first obtains the variance Vr of the pixel values on the multi-valued image IMG (step S220 in FIG. 4). Next, the variance Vr is compared with a predetermined threshold value BR_TH (step S221). As a result of comparing the variance Vr of step S221 with the threshold value BR_TH, Vr <BR_T
If H holds, the character density section is determined to be “blank” (step S223).

If the result of comparing the variance Vr and the threshold value BR_TH in step S221 does not satisfy Vr <BR_TH, the variance Vr is compared with a predetermined threshold value KOSA_TH (step S222).
In step S222, the variance Vr and the threshold value KOSA
If Vr <KOSA_TH holds as a result of comparison with _TH, the character density section is determined to be “light” (step S224). As a result of comparing the variance Vr with the threshold value KOSA_TH in step S222, Vr <KO
If SA_TH does not hold, the character density section is determined to be “dark” (step S225). Then, the character density determining unit 3113 ends the processing of this flowchart, that is, the step S22 of FIG.

Next, the binarizing section 311 initializes the variable KOSA to '0' (step S23 in FIG. 3). Variable K
OSA is a variable for indicating that if KOSA = 0, the character density section is “blank” or “light”, and if KOSA = 1, the character density section is “dark”.

Next, referring to the character density division determined by the character density determination section 3113, it is checked whether the density division is "blank" (step S24). If the density classification is “blank”, the binary image generation unit 3115 determines that the multi-valued image IM
As a binary image of G, a binary image IMB having all pixel values of '0', that is, all white pixels is generated (step S29). If the character density section is not "blank" in step S24, it is next checked whether the character density section is "light" (step S25). If the character density division is “light” in step S25, the binary image generation unit 31
15 is to convert the multi-valued image IMG into the binary threshold value BT.
Generate a binary image IMB binarized with H (step S
28). If the character density category is not “light” in step S25, that is, if the character density category is “dark”, the stain determination threshold value calculation unit 3114 determines whether the connected black pixel on the binary image IMB is dirty or character stroke. A dirt determination threshold value DTH for determining whether the dirt is present is calculated (step S26).

The dirt discriminating threshold value calculating unit 3114 operates as shown in FIG.
As described above, the above-described discriminant analysis method is applied to pixel values on the density histogram that are smaller than the binarization threshold BTH, and pixels that are smaller than the binarization threshold BTH are dark pixels (small pixel values). And a threshold value for separating the pixel into a thin pixel (large pixel value), and this is set as a dirt determination threshold value DTH (step S26). Next, '1' is substituted for the variable KOSA (step S27). Binary image generation unit 3115
Generates a binary image IMB obtained by binarizing the multi-value image IMG with the binary threshold value BTH (step S2).
8). Then, the binarizing unit 311 ends the processing of this flowchart, that is, the step S2 of FIG.

Next, the noise removing unit 312 removes noise from the binary image IMB (step S in FIG. 2).
3). The noise elimination unit 312 first performs the binary image IMB
, All connected black pixels existing in the image are detected (step S30 in FIG. 5). The process for finding a connected black pixel can be realized by, for example, a labeling process which is a known technique. Next, one of the detected connected black pixels is selected and set as a connected black pixel BLK (Step S).
31).

Next, the height H and width W of the connected black pixel BLK are compared with a predetermined threshold value SIZE_S (step S32). The threshold value SIZE_S is a threshold value for determining whether or not to consider the connected black pixel BLK as minute noise from the width W and height H of the connected black pixel BLK. Both the width W and the height H are equal to the threshold value SIZE_
If not more than S, the connected black pixel removing unit 3122 removes the connected black pixel BLK as minute noise (step S3).
6) It is determined whether all the connected black pixels detected in step S30 have been selected (step S37). If it is determined that there is an unselected connected black pixel, the process returns to step S31, selects one unselected connected black pixel, and executes step S32 and subsequent steps.

In step S32, if either the width W or the height H of the connected black pixel BLK is larger than the threshold value SIZE_S, then the width W and the height H and a predetermined threshold value SIZE_M are set. Then, it is checked whether the value of the variable KOSA is “1” (step S33). The threshold value SIZE_M is a threshold value for judging whether or not a certain connected black pixel is regarded as dirty based on the width W and height H of the connected black pixel BLK.

The threshold SIZE_S is set to a value smaller than the threshold SIZE_M.
Either the width W or the height H of the connected black pixel BLK is larger than the threshold SIZE_M or the variable KOSA
Is "0", the process proceeds to step S37, and it is determined whether all the connected black pixels detected in step S30 have been selected. On the other hand, in step S33, both the width W and the height H of the connected black pixel BLK are equal to the threshold SIZE_.
If the difference is equal to or less than M and the value of the variable KOSA is “1”, the stain determination unit 3121 determines whether the connected black pixel BLK is dirty (step S34).

In this case, as shown in FIG. 7, the dirt discriminating unit 3121 controls the variables BK_CNT and TK_CNT.
Are respectively substituted for 0 (step S340 in FIG. 7).
The variable BK_CNT is a variable for storing the number of black pixels constituting the connected black pixel BLK. Also, the variable TK_CN
T is a variable that stores the number of pixels having a pixel value less than the dirt determination threshold value DTH on the multi-value image IMG among the black pixels constituting the connected black pixel BLK.

Next, one of the black pixels constituting the connected black pixel BLK is selected (step S341). Next, the multi-value image I corresponding to the black pixel selected in step S341
The pixel value BK_G of the pixel on the MG is obtained. That is, a multi-valued pixel value before binarization is obtained for the selected black pixel (step S342). Next, the pixel value BK
_G is compared with the dirt determination threshold value DTH (step S343). If BK_G <DTH holds as a result of the comparison in step S343, the variable TK_C
The value of NT is increased by 1 (step S344). If BK_G <DTH does not hold as a result of the comparison in step S343, the process proceeds to step S345. Next, the value of the variable BK_CNT is increased by 1 (step S34).
5).

Next, in step S341, it is determined whether all the constituent black pixels of the connected black pixel BLK have been selected (step S346). If it is determined in step S346 that there is an unselected black pixel, the process returns to step S341, selects one unselected black pixel, and executes step S342 and subsequent steps. If it is determined in step S346 that there is no unselected black pixel, the dirt evaluation value TK_
The value of RTO is calculated (step S347).

TK_RTO = TK_CNT / BK_CNT (1) Next, the dirt evaluation value TK_RTO is compared with a preset dirt evaluation threshold value DTK_TH, and TK_RTO <D
It is determined whether TK_TH holds (step S3)
48). As a result of comparing the dirt evaluation value TK_RTO with a preset dirt evaluation threshold value DTK_TH in step S348, if TK_RTO <DTK_TH holds, it is determined that the connected black pixel BLK is "dirty" (step S349). . If TK_RTO <DTK_TH does not hold as a result of the evaluation in step S348, it is determined that the connected black pixel BLK is a “character stroke” (step S34a). Then, the stain determination unit 3121 ends the process of this flowchart, that is, step S34 in FIG.

Next, the noise removing unit 312 determines in step S
The determination result of the dirt determination unit 3121 at 34 is “dirt”.
Is checked (step S35 in FIG. 5). In step S35, when the determination result of the stain determination unit 3121 is “dirt”, the connected black pixel removal unit 3122 removes the connected black pixel BLK from the binary image IMB (step S36).

Next, the process proceeds to step S37,
At 31, it is determined whether all the connected black pixels on the binary image IMB have been selected. In step S35, when the determination result of the dirt determination unit 3121 is not “dirt”,
Proceeding to step S37, it is determined in step S31 whether all connected black pixels on the binary image IMB have been selected. In step S37, when it is determined that all the connected black pixels on the binary image IMB have been selected,
The noise removing unit 312 ends the processing of this flowchart, that is, the step S3 in FIG. If it is determined that there is an unselected connected black pixel, the process returns to step S31, selects one unselected connected black pixel, and executes step S32 and subsequent steps.

Next, the feature extracting unit 32 extracts features necessary for character recognition from the binary image IMB generated by the preprocessing unit 31 (step S4 in FIG. 2). Next, the character recognition unit 33 compares the features extracted by the feature extraction unit 32 with the character features stored in the character recognition dictionary unit 22 and performs a character recognition process (step S5). Then, the output device 4 outputs the result of character recognition by the character recognition unit 33 as a result of reading the multi-valued image IMG (step S6). Note that the feature extraction unit 32 and the character recognition unit 33 can be realized by applying a known technique.

In the character reader described above, FIGS.
The character reading method shown in FIG. 7 is realized.
In other words, this is a method of reading handwritten characters written on a form, determining a density division which is a division of the density of a character on a multi-valued image obtained by reading an image on the form, and reading the object to be read in the multi-valued image. A character that calculates a stain determination threshold value for determining a character that is a character and stains other than the character, and determines whether a connected black pixel on the binary image obtained by binarizing the multi-valued image is a stain or a character stroke. The reading method has been realized. The above-mentioned stains are unerasable stains or the like generated when erased with an eraser, and the connected black pixels are determined as stains based on the pixel values of the multi-valued image.

As described above, in the character reading device according to the present embodiment, in the binary image obtained by binarizing the multivalued image fetched from the image input device 1 and stored in the image storage unit 21, characters are erased with an eraser. When stains such as unerased parts generated when erasing are forming relatively large connected black pixels,
The dirt determining unit 3121 can determine the connected black pixel as dirt based on the pixel value of the multi-valued image.
Then, by removing such dirt from the binary image, it is possible to avoid performing the character recognition processing by erroneously considering dirt as a part of the character. Therefore, high-precision character reading that is robust against stains such as unerased parts can be realized.

[0043]

Next, the operation of this embodiment will be described in detail with reference to specific examples. Here, the case of reading the numeral "5" as shown in FIG. 8 will be described. Referring to FIG. 8, a reading target described in a form or the like remains when a character component 100 and a character component 101 constituting the number “5” are erased, for example, a character once written with an eraser. Soil component 102 which is a stain
And a dirt component 103. The image of the numeral “5” shown in FIG. 8 is captured by the image input device 1 and stored in the image storage unit 21 as a multi-valued image IMG (step S1 in FIG. 2).

Next, the binarizing section 311 outputs the multi-valued image IM
A process of generating a binary image IMB obtained by binarizing G is performed (step S2). Density histogram generation unit 3111
Generates a density histogram of the multi-valued image IMG (step S20 in FIG. 3). Next, the binarization threshold value calculation unit 3112 uses the multi-valued image I based on the density histogram.
A binarization threshold value BTH for binarizing the MG is calculated (step S21). As a result, it is assumed that BTH = 12.

Next, the character density determination unit 3113 determines the density classification of the character on the multi-valued image IMG (step S2).
2). First, the character density determination unit 3113
The variance Vr of the pixel values of all the pixels on G is obtained (Step S220 in FIG. 4). At this time, it is assumed that the variance Vr is 2.4. Next, it is checked whether or not Vr <BR_TH holds (step S221). BR_TH is a preset threshold value for determining whether or not the multi-valued image IMG is “blank” based on the variance Vr.
It is assumed that _TH is set to 0.2.

Therefore, since Vr <BR_TH does not hold, the process proceeds to step S222. Next, Vr <KO
It is determined whether SA_TH holds (step S2).
22). KOSA_TH is a preset threshold value for determining whether the character density division on the multi-valued image IMG is “dark” or “light” based on the variance Vr.
It is assumed that A_TH = 1.0. Therefore, since Vr <KOSA_TH does not hold, it is determined that the density classification of the character appearing in the multi-value image IMG is “dark”.

Next, the binarizing unit 311 initializes a variable KOSA with 0 (step S23 in FIG. 3). Next, the character density classification determined by the character density determination unit 3113 is “blank”.
It is determined whether it is (Step S24). Here, since the character density category is “dark”, the process proceeds to step S25, and it is determined whether the character density category is “light”. Since the character density category is “dark”, the stain determination threshold value calculation unit 3114 calculates a stain determination threshold value based on the density histogram (step S26).

That is, since BTH = 12, FIG.
The above-described discriminant analysis method is applied to the portion where the pixel value of the density histogram shown in FIG. Here, it is assumed that DTH = 10 is obtained. Next, the binarizing unit 311 substitutes 1 for a variable KOSA (Step S27). Then, the binary image generation unit 3115 converts the multi-valued image IMG into a binary threshold BT.
H is used to generate a binary image IMB shown in FIG. 9 (step S28). If it is determined in step S24 that the character density section is “blank” in the determination of “blank”, the binary image IMB in which all pixels are white pixels is determined as the binary image of the multi-level image IMG. Generate

Next, the noise removing unit 312 detects a connected black pixel from the binary image IMB (step S30 in FIG. 5). As a result, four connected black pixels 110 to 113 shown in FIG. 9 are detected. Among these four connected black pixels, first, the connected black pixel 110 is selected (step S31). At this time, the size of the connected black pixel 110 is
It is assumed that the width W is 20 pixels and the height H is 35 pixels.

Next, it is determined whether or not both the width W and the height H are equal to or smaller than a preset threshold value SIZE_S (step S32). Here, the threshold value SIZE_S
= 5. In this case, since both the width W and the height H are larger than SIZE_S,
Then, it is determined whether or not the height H is equal to or smaller than a preset threshold value SIZE_M and the variable KOSA = 1 (step S33). Here, the threshold value SIZE_M
= 10 is assumed to be set.

In this case, since both the width W and the height H are larger than the threshold value SIZE_M, the flow advances to step S37 to determine whether all the connected black pixels have been selected. However, since the connected black pixels 111 to 113 have not been selected yet, the process returns to step S31, and the connected black pixel 111 is selected. The size of the connected black pixel is as follows: width W = 10, height H = 4
Assume that Then, in the size determination in step S32, since both the width W and the height H do not satisfy the condition that the width W and the height H are equal to or smaller than the threshold SIZE_S, the process proceeds to step S33. In the size determination in step S33, the width W and the height H
Satisfy the condition that both are equal to or smaller than SIZE_M, the stain determination unit 3121 determines whether or not the connected black pixel 111 is dirty (step S34).

Next, the dirt determining unit 3121 sets the variable BK_
Both CNT and the variable TK_CNT are initialized to 0 (FIG. 7)
Step S340). Next, one of the black pixels constituting the connected black pixel 111 is selected (step S34).
1). Next, as shown in FIG. 10, the pixel value BK_G of the pixel corresponding to the selected black pixel is stored in the image storage unit 2.
It is determined by referring to the multi-valued image IMG stored in No. 1 (step S342). FIG. 10 shows a connected black image BLK on the binary image IMB and a multi-value image IMG.

When determining the pixel value BK_G of the pixel corresponding to the selected black pixel, reference is made to the character component 101 of FIG. 8 corresponding to the connected black pixel 111 shown in FIG.
Here, it is assumed that BK_G = 4. Next, this B
K_G is compared with the dirt determination threshold DTH calculated by the dirt determination threshold calculator 3114, and BK_G <D
It is determined whether or not TH is satisfied (step S34)
3). As described above, the dirt determination threshold DTH is DTH =
10, BK_G <D
TH holds. Therefore, the value of the variable TK_CNT is increased by 1 (step S344). And the variable BK_C
The value of NT is increased by 1 (step S345).

Next, it is checked whether or not all the black pixels have been selected (step S346). Here, assuming that there is still an unselected black pixel, the process returns to step S341, and one unselected black pixel is selected. Then, the pixel value of the pixel on the multi-valued image IMG corresponding to the black pixel is determined as BK_G (step S342). Here, this pixel value BK_G
= 12. Next, the pixel value BK_G is compared with the dirt determination threshold value DTH to determine whether BK_G <DTH holds (step S343). But,
Since BK_G <DTH does not hold, step S34 is performed.
4 is skipped and the process advances to step S345 to set the variable BK_CNT.
Is incremented by one.

Then, in a step S346, it is determined whether or not all the black pixels have been selected. In this manner, steps S341 to S34 are performed until all the black pixels constituting the connected black pixel 111 are selected in step S341.
The processing between 6 is repeated. When all the black pixels have been selected, the process proceeds to step S347, and the value of TK_RTO is calculated. Here, TK_CNT = 20, BK_CN
If T = 25, TK_RTO = 0.8. Next, TK_RTO is compared with a preset threshold value DTK_TH. Here, DTK_TH =
It is assumed that it is set to 0.1. Then, step S
As a result of comparing TK_RTO and DTK_TH in 348, since TK_RTO <DTK_TH does not hold,
It is determined that the connected black pixel 111 is a “character stroke” (step S34a).

Then, the process proceeds to step S35 in FIG. 5, and it is determined whether or not the determination result of the dirt determining unit 3121 is "dirt". In this case, the stain determination unit 3121 determines that the connected black pixel 11
Since 1 is determined to be a “character stroke”, the process proceeds to step S37. In step S37, it is determined whether all the connected black pixels have been selected. In this case, since the connected black pixel 112 and the connected black pixel 113 have not been selected, the process returns to step S31, and the connected black pixel 112 is selected.

The size of the connected black pixel 112 is W = 5,
Assume that the height H = 8. Then, in the determination in step S32, since the condition that both the width W and the height H do not satisfy the condition that the width W and the height H are equal to or smaller than SIZE_S is satisfied, the process proceeds to step S33. In the determination in step S33, the width W and the height H
Are smaller than or equal to SIZE_M, and the variable KOSA = 1
Is satisfied, the stain determination unit 3121 determines that the connected black pixel 11
It is determined whether 2 is dirty (step S34).

Next, the stain determination unit 3121 performs the processing from step S340 to step S346 in FIG. As a result, TK_CNT = 2,
If BK_CNT = 32 is obtained, TK_RTO
= 0.0625. Next, this TK_RTO is compared with the threshold value DTK_TH (step S34).
8). Here, DTK_TH = 0.1, so that in step S348, TK_RTO <DTK_TH
Holds, it is determined that the connected black pixel 112 is “dirty” (step S349). Then, the process proceeds to step S35 in FIG. 5, and it is determined whether or not the determination result of the dirt determining unit 3121 is “dirty”. In this case, the stain determination unit 3121
Has determined that the connected black pixel 112 is 'dirty', the connected black pixel removing unit 3122 removes the connected black pixel 112 from the binary image IMB (step S36).

Next, in step S37, it is determined whether all the connected black pixels have been selected. Since the connected black pixel 113 has not been selected, the process returns to step S31, and the connected black pixel 11
Select 3. The size of the connected black pixel 113 is represented by the width W =
3, height H = 4. Then, step S32
Since the width W and the height H are both equal to or smaller than SIZE_S in the determination of (1), the connected black pixel removing unit 3122 removes the connected black pixel 113 from the binary image IMB (step S36). Next, in step S37, it is determined whether all the connected black pixels have been selected. At this time, since all of the connected black pixels 110 to 113 on the binary image IMB have already been selected, the noise removing unit 312 ends the processing (Step S3 in FIG. 2).

Next, the feature extraction unit 32 receives the noise-removed binary image IMB shown in FIG. 11 from the pre-processing unit 31 and extracts features used when the character recognition unit 33 performs the character recognition processing ( Step S4 in FIG. 2).

Next, the character recognizing unit 33 includes the feature extracting unit 32
Are compared with the character features stored in the character recognition dictionary unit 22 to perform a character recognition process (step S).
5). Then, the category name to which the multi-value image IMG belongs, which is obtained as a result of the character recognition processing, is output to the output device 4 as a character reading result.

In the embodiment described above, the functions from the preprocessing unit 31 to the character recognition unit 33 are realized by the data processing device 3 operating under program control executing the program stored in the memory. It had been. On the other hand, as shown in FIG. 12, a recording medium 5 storing a program to be executed by the data processing device 3 is provided.
A program may be read from the recording medium 5 into the memory of the data processing device 3 to control the operation of the data processing device 3.

That is, by preparing a recording medium on which a program for realizing the processes of FIGS. 2 to 4, 6 and 7 described above is prepared, and using this to control each part of FIG. 12,
Obviously, a character reading operation similar to that described above can be performed. As the recording medium 5, various recording media other than the semiconductor memory and the magnetic disk device can be used.

If the computer is controlled by the program recorded on the recording medium, it is apparent that the character reading operation can be performed in the same manner as described above.
As this recording medium, various recording media other than the semiconductor memory and the magnetic disk device can be used.

[0065]

As described above, the present invention refers to density information on a multi-valued image before binarization, to determine whether a certain large connected black pixel on the binary image is a stain or a character stroke. As a result, it is possible to remove only the dirt without mistakenly removing the character stroke, and it is stable against stains such as unerased characters that occur when characters written once with the eraser are erased. There is an effect that a character reading device can be realized.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of an embodiment of a character reading device according to the present invention.

FIG. 2 is a flowchart showing the operation of each unit in FIG.

FIG. 3 is a flowchart showing an operation of a binarizing unit 311 in FIG.

FIG. 4 is a flowchart illustrating an operation of a character density determination unit 3113 in FIG. 1;

FIG. 5 is a diagram illustrating an example of a density histogram generated by a density histogram generation unit 3111 in FIG. 1;

FIG. 6 is a flowchart showing an operation of a noise removing unit 312 in FIG.

FIG. 7 is a flowchart showing an operation of a stain determination unit 3121 in FIG.

FIG. 8 is a diagram illustrating an example of a multi-valued image IMG.

FIG. 9 is a diagram illustrating an example of a binary image IMB obtained by binarizing a multi-value image IMG.

FIG. 10 is a diagram illustrating a process of referring to a pixel value BK_G of a pixel on the multi-value image IMG corresponding to the pixel from a pixel on the connected black pixel BLK on the binary image IMB.

11 is a diagram illustrating an example of a binary image IMB after noise removal by a noise removal unit 312 in FIG. 1;

FIG. 12 is a block diagram showing another embodiment of the character reading device according to the present invention.

[Explanation of symbols]

 Reference Signs List 1 image input device 2 storage device 21 image storage unit 22 character recognition dictionary unit 3 data processing unit 31 preprocessing unit 311 binarization unit 3111 density histogram generation unit 3112 binarization threshold calculation unit 3113 character density determination unit 3114 dirt discrimination Threshold calculator 3115 Binary image generator 312 Noise remover 3121 Stain determiner 3122 Concatenated black pixel remover 32 Feature extractor 33 Character recognizer 4 Output device 5 Recording medium 100, 101 Character component 102, 103 Dirt component 110 , 111,112,113 connected black pixel

Claims (9)

[Claims] 1. A character reading apparatus for reading handwritten characters written on a form, wherein a character for determining a density category is determined for a character on a multi-valued image obtained by reading an image on the form. A density determination unit, a stain determination threshold value calculation unit that calculates a stain determination threshold value for determining a character to be read in the multi-valued image and a stain other than the character, A character reading device, comprising: a dirt determining unit that determines whether a connected black pixel on the binarized binary image is a dirt or a character stroke. 2. The character reading device according to claim 1, wherein the stain is a stain such as an unerased part generated when the stain is erased with an eraser. 3. The character reading device according to claim 1, wherein the stain determining unit determines the connected black pixel as a stain based on a pixel value of the multi-valued image. 4. A character reading method for reading a handwritten character written on a form, wherein a character on a multi-valued image obtained by reading an image on the form determines a density classification which is a density classification. A density determination step; a dirt determination threshold value calculation step of calculating a dirt determination threshold value for determining a character to be read in the multi-valued image and a dirt other than the character; A stain determining step of determining whether a connected black pixel on the binarized binary image is a stain or a character stroke. 5. The character reading method according to claim 4, wherein the stain is an unerasable stain or the like generated when the stain is erased with an eraser. 6. The character reading method according to claim 4, wherein in the dirt determining step, the connected black pixels are determined as dirt based on pixel values of the multi-valued image. 7. A recording medium on which a computer reads a character reading program for reading handwritten characters written on a form, the program comprising: A character density determining step of determining a density category which is a density category of a character on an image; and a stain determination threshold value for determining a character to be read in the multi-valued image and a stain other than the character. And a stain determining step of determining whether a connected black pixel on the binary image obtained by binarizing the multi-valued image is a stain or a character stroke. Medium. 8. The recording medium according to claim 7, wherein the stain is an unerasable stain or the like generated when the stain is erased with an eraser. 9. The recording medium according to claim 7, wherein in the stain determination step, the connected black pixel is determined as stain based on a pixel value of the multi-valued image.